chore: support accelerator id in the column

api/v1alpha1/nodeclaim_types.go

@@ -41,6 +41,17 @@ type NodeClaimSpec struct {
 	// +optional
 	Region string `json:"region,omitempty"`
 
+	// AcceleratorID is the provider's own identifier for what actually serves this
+	// claim (e.g. AWS "p5.48xlarge", RunPod "NVIDIA H100 80GB HBM3"), resolved at
+	// placement time from the Pod's requested accelerator type + count via the
+	// provider's MapAccelerator. Unlike Provider/CapacityType/Region it is NOT a
+	// provisioning input (the provider re-derives it from the Pod) — it is recorded
+	// for reporting, like PoolRef, so `kubectl get nc` shows the concrete SKU behind
+	// each instance without cross-referencing the Pod. Empty for a CPU-only claim,
+	// which requests no accelerator.
+	// +optional
+	AcceleratorID string `json:"acceleratorID,omitempty"`
+
 	// PoolRef is the NodePool whose policy produced this claim, for reporting.
 	// +optional
 	PoolRef string `json:"poolRef,omitempty"`
@@ -139,9 +150,10 @@ type NodeClaimStatus struct {
 // +kubebuilder:subresource:status
 // +kubebuilder:printcolumn:name="Provider",type=string,JSONPath=`.spec.provider`
 // +kubebuilder:printcolumn:name="Region",type=string,JSONPath=`.spec.region`
-// +kubebuilder:printcolumn:name="Instance",type=string,JSONPath=`.status.instanceID`
-// +kubebuilder:printcolumn:name="CapacityType",type=string,JSONPath=`.spec.capacityType`
+// +kubebuilder:printcolumn:name="ACCELERATOR_ID",type=string,JSONPath=`.spec.acceleratorID`
+// +kubebuilder:printcolumn:name="CAPACITY_TYPE",type=string,JSONPath=`.spec.capacityType`
 // +kubebuilder:printcolumn:name="Phase",type=string,JSONPath=`.status.phase`
+// +kubebuilder:printcolumn:name="Instance",type=string,JSONPath=`.status.instanceID`
 // +kubebuilder:printcolumn:name="Age",type=date,JSONPath=`.metadata.creationTimestamp`
 
 // NodeClaim represents one external GPU instance and its lifecycle.

config/crd/bases/nebula.inftyai.com_nodeclaims.yaml

@@ -23,15 +23,18 @@ spec:
     - jsonPath: .spec.region
       name: Region
       type: string
-    - jsonPath: .status.instanceID
-      name: Instance
+    - jsonPath: .spec.acceleratorID
+      name: ACCELERATOR_ID
       type: string
     - jsonPath: .spec.capacityType
-      name: CapacityType
+      name: CAPACITY_TYPE
       type: string
     - jsonPath: .status.phase
       name: Phase
       type: string
+    - jsonPath: .status.instanceID
+      name: Instance
+      type: string
     - jsonPath: .metadata.creationTimestamp
       name: Age
       type: date
@@ -67,6 +70,17 @@ spec:
               ledger, not a spec: its reason to exist is to survive the Node so teardown can
               reclaim the external instance and never leak a paid GPU.
             properties:
+              acceleratorID:
+                description: |-
+                  AcceleratorID is the provider's own identifier for what actually serves this
+                  claim (e.g. AWS "p5.48xlarge", RunPod "NVIDIA H100 80GB HBM3"), resolved at
+                  placement time from the Pod's requested accelerator type + count via the
+                  provider's MapAccelerator. Unlike Provider/CapacityType/Region it is NOT a
+                  provisioning input (the provider re-derives it from the Pod) — it is recorded
+                  for reporting, like PoolRef, so `kubectl get nc` shows the concrete SKU behind
+                  each instance without cross-referencing the Pod. Empty for a CPU-only claim,
+                  which requests no accelerator.
+                type: string
               capacityType:
                 description: |-
                   CapacityType is the purchase tier the placement optimizer selected

config/samples/deployment.yaml

@@ -23,7 +23,7 @@
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: gpu-workload-sample-l4
+  name: gpu-workload-sample
   namespace: default
   labels:
     app.kubernetes.io/managed-by: nebula
@@ -37,8 +37,8 @@ spec:
       labels:
         app: gpu-workload-sample
         nebula.inftyai.com/enabled: "true"
-        nebula.inftyai.com/nodepool: aws
-        nebula.inftyai.com/accelerator-type: l4
+        nebula.inftyai.com/nodepool: sample
+        nebula.inftyai.com/accelerator-type: a100-40gb
     spec:
       # Do NOT set nodeName or a provider nodeSelector yourself — the placement
       # controller fills the nodeSelector in when it ungates the Pod. Setting
@@ -65,4 +65,4 @@ spec:
             # GPU count. Standard extended resource, so the scheduler's fit check
             # and provisioning read the same number. 8 => 8x the accelerator-type
             # above.
-            nvidia.com/gpu: "1"
+            nvidia.com/gpu: "8"

config/samples/nebula_v1alpha1_nodepool_neocloud.yaml → config/samples/nebula_v1alpha1_nodepool.yaml

@@ -3,16 +3,25 @@ kind: NodePool
 metadata:
   labels:
     app.kubernetes.io/managed-by: nebula
-  name: neocloud
+  name: sample
 spec:
-  # The NeoClouds this pool may place onto. Order matters only for the Ordered
   # strategy (the inner, provider-ranking axis).
   providers:
   - name: modal
+  - name: aws
+    regions:
+    - us-east-1
+    - us-west-1
+    - ap-south-1
+    - ap-northeast-1
+    - eu-central-1
+    - eu-west-1
+    - ca-central-1
+    - sa-east-1
+  - name: modal
   # - name: runpod
-  # Outer axis: try Reserved on every provider first, fall back to on-demand, then spot.
+  # Outer axis: try OnDemand on every provider first, fall back to Spot.
   capacityTypes:
-  - Reserved
   - OnDemand
   - Spot
   # Inner axis: within the active capacity tier.

internal/controller/pod_placement_controller.go

@@ -186,6 +186,10 @@ type placement struct {
 	// Empty means the provider's configured default region; region-simple
 	// providers leave it empty.
 	region string
+	// acceleratorID is the provider's own identifier for what serves this request
+	// (e.g. AWS "p5.48xlarge"), resolved via MapAccelerator(type, count). Empty for
+	// a CPU-only Pod, which requests no accelerator.
+	acceleratorID string
 }
 
 // needsPlacement reports whether the Pod is an opted-in workload still held by

internal/controller/pod_placement_controller_test.go

@@ -181,6 +181,11 @@ func TestPlacement_UngatesAndRoutesAndCreatesClaim(t *testing.T) {
 	if nc.Spec.Provider != provider.ProviderModal || nc.Spec.PodRef.UID != "uid-1" || nc.Spec.PoolRef != "pool-a" {
 		t.Fatalf("unexpected claim spec: %+v", nc.Spec)
 	}
+	// The provider's resolved accelerator id is recorded for reporting (the fake's
+	// MapAccelerator echoes the canonical type, so H100 -> "H100").
+	if nc.Spec.AcceleratorID != "H100" {
+		t.Fatalf("expected claim to record acceleratorID H100, got %q", nc.Spec.AcceleratorID)
+	}
 }
 
 func TestPlacement_FirstMatchingProviderWins(t *testing.T) {
@@ -246,6 +251,14 @@ func TestPlacement_CPUOnlyPodMatchesAnyProvider(t *testing.T) {
 	if got.Spec.NodeSelector[nebulav1alpha1.ProviderLabel] != provider.ProviderModal {
 		t.Fatalf("expected a CPU-only Pod to place on modal, got %v", got.Spec.NodeSelector)
 	}
+	// A CPU-only claim requests no accelerator, so the id is left empty.
+	var nc nebulav1alpha1.NodeClaim
+	if err := c.Get(context.Background(), types.NamespacedName{Name: "default-p1"}, &nc); err != nil {
+		t.Fatalf("expected NodeClaim default-p1: %v", err)
+	}
+	if nc.Spec.AcceleratorID != "" {
+		t.Fatalf("expected empty acceleratorID for a CPU-only claim, got %q", nc.Spec.AcceleratorID)
+	}
 }
 
 func TestPlacement_SkipsPodWithoutOptInLabel(t *testing.T) {

internal/controller/pod_placement_helpers.go

@@ -139,9 +139,10 @@ func (r *PodPlacementReconciler) selectPlacement(ctx context.Context, pod *corev
 				log.Info("selected placement candidate",
 					"provider", ref.Name, "capacityType", tier, "region", region)
 				return placement{
-					provider:     ref.Name,
-					capacityType: tier,
-					region:       region,
+					provider:      ref.Name,
+					capacityType:  tier,
+					region:        region,
+					acceleratorID: acceleratorID,
 				}, true, 0
 			}
 		}
@@ -234,10 +235,11 @@ func (r *PodPlacementReconciler) ensureClaim(ctx context.Context, pod *corev1.Po
 				Name:      pod.Name,
 				UID:       string(pod.UID),
 			},
-			Provider:     p.provider,
-			CapacityType: p.capacityType,
-			Region:       p.region,
-			PoolRef:      pool.Name,
+			Provider:      p.provider,
+			CapacityType:  p.capacityType,
+			Region:        p.region,
+			AcceleratorID: p.acceleratorID,
+			PoolRef:       pool.Name,
 		},
 	}
 	err = r.Create(ctx, claim)